Fwd: [torqueusers] Confused about vmem

Diego M. Vadell dvadell at linuxclusters.com.ar
Sun Oct 5 17:25:49 MDT 2008


Hi Tom

   Thank you very much for your answer. Sadly, I am very sure I did the 
memtest in parallel. In fact, I did all the test twice.

Thanks
 -- Diego

On Friday 03 October 2008 17:01:42 Tom Rudwick wrote:
> I think you want vmem=4G, so the total
> memory of all tasks will be 4GB.
>
> I don't know if it works correctly.
> In your test 2, are you sure the memtests
> were simultaneous? Or could they have been
> done one after the other?
>
> Tom
>
> Diego M. Vadell wrote:
> > Anyone? Any hint will be appreciated.
> >
> > TIA
> >  -- Diego
> >
> >
> > ----------  Forwarded Message  ----------
> >
> > Subject: [torqueusers] Confused about vmem
> > Date: Monday 29 September 2008
> > From: "Diego M. Vadell" <dvadell at linuxclusters.com.ar>
> > To: torqueusers at supercluster.org
> >
> > Hi list,
> >
> >    We are starting to restrict the amount of RAM with torque. We have 4
> > cores nodes with 4Gb of RAM.
> >
> >    I made some tests with setting, via qmgr, a default of pvmem = 1Gb and
> > it seemed to be what we were looking for, until we considered a
> > particular job.
> >
> >    This job uses openMP. It is one process that needs a whole node: the
> > 4Gb and the 4 cores.
> >
> >    If I ask for nodes=1:ppn=4,pvmem=1G, torque will get me a whole node,
> > but it will kill this job's only process  when it uses more than 1G. If I
> > ask for nodes=1:ppn=4,pvmem=4G, torque will look for a node with 16Gb of
> > RAM, and I don't have one of those.
> >
> >    So I found vmem. As I can't make lots of test right now (I have to
> > schedule downtime to do that safely) I made some little ones (with a
> > program called memtest that just malloc's and fills memory):
> >
> > 1) Got an interactive session with vmem=1Gb and ran 1 memtest:
> > qsub -I -l nodes=1:ppn=2 -l vmem=1000mb
> > Commmand line: examples/memtest
> > Result: malloc failure after 993 MiB
> > Expected: This is Ok as vmem is the total amount of virtual memory the
> > whole job can use.
> >
> > 2) Got an interactive session with vmem=1Gb and ran 2 memtest, in
> > parallel: qsub -I -l nodes=1:ppn=2 -l vmem=1000mb
> > Command line: examples/memtest & examples/memtest
> > Result: malloc failure after 993 MiB
> >            malloc failure after 993 MiB
> > Expected: So both can allocate almost 1Gb! I was expecting them to die at
> > 500Mb, so they would sum the max 1Gb I asked torque.
> >
> > 3) Got an interactive session with pvmem=1Gb and ran 2 memtest, in
> > parallel: qsub -I -l nodes=1:ppn=2 -l pvmem=1000mb
> > Command line: examples/memtest & examples/memtest
> > Result: malloc failure after 993 MiB
> >            malloc failure after 993 MiB
> > Expected: This is what I was expecting because both can allocate about
> > 1Gb.
> >
> > So vmem and pvmem seem to do the same. What am I missing here?
> >
> > Thanks in advance,
> >  -- Diego.
> > _______________________________________________
> > torqueusers mailing list
> > torqueusers at supercluster.org
> > http://www.supercluster.org/mailman/listinfo/torqueusers
> >
> >
> > -------------------------------------------------------
> > _______________________________________________
> > torqueusers mailing list
> > torqueusers at supercluster.org
> > http://www.supercluster.org/mailman/listinfo/torqueusers




More information about the torqueusers mailing list